NL2SQL System

Natural Language to SQL query generation system with human-in-the-loop learning.

Quick Start

pip install -r requirements.txt
echo "OPENAI_API_KEY=sk-your-key-here" > .env
python main.py

Open http://127.0.0.1:7861

Features

SQL Generation

• Natural language to SQL conversion
• Automatic database detection from 29+ Spider databases
• Foreign key relationship awareness
• Human-in-the-loop query refinement

User Interface

• Glassmorphism design
• Real-time query execution
• Interactive SQL editing
• Query regeneration and improvement saving

Learning System

• Vector-based few-shot retrieval
• User correction storage
• Continuous accuracy improvement

Architecture

Question → Database Router → Schema Extraction
    ↓
Table Selection → Few-Shot Retrieval
    ↓
SQL Generation → Human Review → Execution
    ↓
Save Improvement → Vectorstore Update

Project Structure

main.py                  # Application (2,144 lines)
requirements.txt         # Dependencies
spider/                  # Spider dataset
  database/              # 200+ SQLite databases
  train_spider.json
  dev.json
  tables.json
chroma_db/              # Vectorstore
interaction_log.jsonl   # Query logs

Configuration

Edit main.py:

USE_OPENAI = True
OPENAI_MODEL = "gpt-4o-mini"
TEMPERATURE = 0.0

Requirements

• Python 3.11+
• OpenAI API key
• 4GB+ RAM
• Internet connection

Dependencies

langchain>=0.2.0
langchain-openai>=0.1.0
chromadb
sqlalchemy
pandas>=2.2.0
gradio
python-dotenv

Usage

Ask natural language questions:

"Show all singers"
"How many stadiums are there?"
"List students with GPA above 3.5"

Click Run Query to execute. Edit SQL if needed. Save improvements for future queries.

Troubleshooting

Issue	Solution
Port in use	Change server_port in main.py
Import errors	pip install -r requirements.txt
API key error	Create .env with OPENAI_API_KEY
Database not found	Type: Use database

License

MIT License

Contact

GitHub: @ksaravindakashyap Repository: github.com/ksaravindakashyap/nl2sql